// Copyright 2023 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.

#include "common/arch.h"
#include "common/archives.h"
#include "common/microprofile.h"
#include "common/scope_exit.h"
#include "common/settings.h"
#include "core/core.h"
#include "core/memory.h"
#include "video_core/debug_utils/debug_utils.h"
#include "video_core/pica/pica_core.h"
#include "video_core/pica/vertex_loader.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/shader/shader.h"

namespace Pica {

MICROPROFILE_DEFINE(GPU_Drawing, "GPU", "Drawing", MP_RGB(50, 50, 240));

using namespace DebugUtils;

union CommandHeader {
    u32 hex;
    BitField<0, 16, u32> cmd_id;
    BitField<16, 4, u32> parameter_mask;
    BitField<20, 8, u32> extra_data_length;
    BitField<31, 1, u32> group_commands;
};
static_assert(sizeof(CommandHeader) == sizeof(u32), "CommandHeader has incorrect size!");

PicaCore::PicaCore(Memory::MemorySystem& memory_, std::shared_ptr<DebugContext> debug_context_)
    : memory{memory_}, debug_context{std::move(debug_context_)}, geometry_pipeline{regs.internal,
                                                                                   gs_unit,
                                                                                   gs_setup},
      shader_engine{CreateEngine(Settings::values.use_shader_jit.GetValue())} {
    InitializeRegs();

    const auto submit_vertex = [this](const AttributeBuffer& buffer) {
        const auto add_triangle = [this](const OutputVertex& v0, const OutputVertex& v1,
                                         const OutputVertex& v2) {
            rasterizer->AddTriangle(v0, v1, v2);
        };
        const auto vertex = OutputVertex(regs.internal.rasterizer, buffer);
        primitive_assembler.SubmitVertex(vertex, add_triangle);
    };

    gs_unit.SetVertexHandlers(submit_vertex, [this]() { primitive_assembler.SetWinding(); });
    geometry_pipeline.SetVertexHandler(submit_vertex);

    primitive_assembler.Reconfigure(PipelineRegs::TriangleTopology::List);
}

PicaCore::~PicaCore() = default;

void PicaCore::InitializeRegs() {
    auto& framebuffer_top = regs.framebuffer_config[0];
    auto& framebuffer_sub = regs.framebuffer_config[1];

    // Set framebuffer defaults from nn::gx::Initialize
    framebuffer_top.address_left1 = 0x181E6000;
    framebuffer_top.address_left2 = 0x1822C800;
    framebuffer_top.address_right1 = 0x18273000;
    framebuffer_top.address_right2 = 0x182B9800;
    framebuffer_sub.address_left1 = 0x1848F000;
    framebuffer_sub.address_left2 = 0x184C7800;

    framebuffer_top.width.Assign(240);
    framebuffer_top.height.Assign(400);
    framebuffer_top.stride = 3 * 240;
    framebuffer_top.color_format.Assign(PixelFormat::RGB8);
    framebuffer_top.active_fb = 0;

    framebuffer_sub.width.Assign(240);
    framebuffer_sub.height.Assign(320);
    framebuffer_sub.stride = 3 * 240;
    framebuffer_sub.color_format.Assign(PixelFormat::RGB8);
    framebuffer_sub.active_fb = 0;

    // Tales of Abyss expects this register to have the following default values.
    auto& gs = regs.internal.gs;
    gs.max_input_attribute_index.Assign(1);
    gs.shader_mode.Assign(ShaderRegs::ShaderMode::VS);
}

void PicaCore::BindRasterizer(VideoCore::RasterizerInterface* rasterizer) {
    this->rasterizer = rasterizer;
}

void PicaCore::SetInterruptHandler(Service::GSP::InterruptHandler& signal_interrupt) {
    this->signal_interrupt = signal_interrupt;
}

void PicaCore::ProcessCmdList(PAddr list, u32 size) {
    // Initialize command list tracking.
    const u8* head = memory.GetPhysicalPointer(list);
    cmd_list.Reset(list, head, size);

    while (cmd_list.current_index < cmd_list.length) {
        // Align read pointer to 8 bytes
        if (cmd_list.current_index % 2 != 0) {
            cmd_list.current_index++;
        }

        // Read the header and the value to write.
        const u32 value = cmd_list.head[cmd_list.current_index++];
        const CommandHeader header{cmd_list.head[cmd_list.current_index++]};

        // Write to the requested PICA register.
        WriteInternalReg(header.cmd_id, value, header.parameter_mask);

        // Write any extra paramters as well.
        for (u32 i = 0; i < header.extra_data_length; ++i) {
            const u32 cmd = header.cmd_id + (header.group_commands ? i + 1 : 0);
            const u32 extra_value = cmd_list.head[cmd_list.current_index++];
            WriteInternalReg(cmd, extra_value, header.parameter_mask);
        }
    }
}

void PicaCore::WriteInternalReg(u32 id, u32 value, u32 mask) {
    if (id >= RegsInternal::NUM_REGS) {
        LOG_ERROR(
            HW_GPU,
            "Commandlist tried to write to invalid register 0x{:03X} (value: {:08X}, mask: {:X})",
            id, value, mask);
        return;
    }

    // Expand a 4-bit mask to 4-byte mask, e.g. 0b0101 -> 0x00FF00FF
    constexpr std::array<u32, 16> ExpandBitsToBytes = {
        0x00000000, 0x000000ff, 0x0000ff00, 0x0000ffff, 0x00ff0000, 0x00ff00ff,
        0x00ffff00, 0x00ffffff, 0xff000000, 0xff0000ff, 0xff00ff00, 0xff00ffff,
        0xffff0000, 0xffff00ff, 0xffffff00, 0xffffffff,
    };

    // TODO: Figure out how register masking acts on e.g. vs.uniform_setup.set_value
    const u32 old_value = regs.internal.reg_array[id];
    const u32 write_mask = ExpandBitsToBytes[mask];
    regs.internal.reg_array[id] = (old_value & ~write_mask) | (value & write_mask);

    // Track register write.
    DebugUtils::OnPicaRegWrite(id, mask, regs.internal.reg_array[id]);

    // Track events.
    if (debug_context) {
        debug_context->OnEvent(DebugContext::Event::PicaCommandLoaded, &id);
        SCOPE_EXIT({ debug_context->OnEvent(DebugContext::Event::PicaCommandProcessed, &id); });
    }

    switch (id) {
    // Trigger IRQ
    case PICA_REG_INDEX(trigger_irq):
        signal_interrupt(Service::GSP::InterruptId::P3D);
        break;

    case PICA_REG_INDEX(pipeline.triangle_topology):
        primitive_assembler.Reconfigure(regs.internal.pipeline.triangle_topology);
        break;

    case PICA_REG_INDEX(pipeline.restart_primitive):
        primitive_assembler.Reset();
        break;

    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.index):
        immediate.Reset();
        break;

    // Load default vertex input attributes
    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.set_value[0]):
    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.set_value[1]):
    case PICA_REG_INDEX(pipeline.vs_default_attributes_setup.set_value[2]):
        SubmitImmediate(value);
        break;

    case PICA_REG_INDEX(pipeline.gpu_mode):
        // This register likely just enables vertex processing and doesn't need any special handling
        break;

    case PICA_REG_INDEX(pipeline.command_buffer.trigger[0]):
    case PICA_REG_INDEX(pipeline.command_buffer.trigger[1]): {
        const u32 index = static_cast<u32>(id - PICA_REG_INDEX(pipeline.command_buffer.trigger[0]));
        const PAddr addr = regs.internal.pipeline.command_buffer.GetPhysicalAddress(index);
        const u32 size = regs.internal.pipeline.command_buffer.GetSize(index);
        const u8* head = memory.GetPhysicalPointer(addr);
        cmd_list.Reset(addr, head, size);
        break;
    }

    // It seems like these trigger vertex rendering
    case PICA_REG_INDEX(pipeline.trigger_draw):
    case PICA_REG_INDEX(pipeline.trigger_draw_indexed): {
        const bool is_indexed = (id == PICA_REG_INDEX(pipeline.trigger_draw_indexed));
        DrawArrays(is_indexed);
        break;
    }

    case PICA_REG_INDEX(gs.bool_uniforms):
        gs_setup.WriteUniformBoolReg(regs.internal.gs.bool_uniforms.Value());
        break;

    case PICA_REG_INDEX(gs.int_uniforms[0]):
    case PICA_REG_INDEX(gs.int_uniforms[1]):
    case PICA_REG_INDEX(gs.int_uniforms[2]):
    case PICA_REG_INDEX(gs.int_uniforms[3]): {
        const u32 index = (id - PICA_REG_INDEX(gs.int_uniforms[0]));
        gs_setup.WriteUniformIntReg(index, regs.internal.gs.GetIntUniform(index));
        break;
    }

    case PICA_REG_INDEX(gs.uniform_setup.set_value[0]):
    case PICA_REG_INDEX(gs.uniform_setup.set_value[1]):
    case PICA_REG_INDEX(gs.uniform_setup.set_value[2]):
    case PICA_REG_INDEX(gs.uniform_setup.set_value[3]):
    case PICA_REG_INDEX(gs.uniform_setup.set_value[4]):
    case PICA_REG_INDEX(gs.uniform_setup.set_value[5]):
    case PICA_REG_INDEX(gs.uniform_setup.set_value[6]):
    case PICA_REG_INDEX(gs.uniform_setup.set_value[7]): {
        gs_setup.WriteUniformFloatReg(regs.internal.gs, value);
        break;
    }

    case PICA_REG_INDEX(gs.program.set_word[0]):
    case PICA_REG_INDEX(gs.program.set_word[1]):
    case PICA_REG_INDEX(gs.program.set_word[2]):
    case PICA_REG_INDEX(gs.program.set_word[3]):
    case PICA_REG_INDEX(gs.program.set_word[4]):
    case PICA_REG_INDEX(gs.program.set_word[5]):
    case PICA_REG_INDEX(gs.program.set_word[6]):
    case PICA_REG_INDEX(gs.program.set_word[7]): {
        u32& offset = regs.internal.gs.program.offset;
        if (offset >= 4096) {
            LOG_ERROR(HW_GPU, "Invalid GS program offset {}", offset);
        } else {
            gs_setup.program_code[offset] = value;
            gs_setup.MarkProgramCodeDirty();
            offset++;
        }
        break;
    }

    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[0]):
    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[1]):
    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[2]):
    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[3]):
    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[4]):
    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[5]):
    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[6]):
    case PICA_REG_INDEX(gs.swizzle_patterns.set_word[7]): {
        u32& offset = regs.internal.gs.swizzle_patterns.offset;
        if (offset >= gs_setup.swizzle_data.size()) {
            LOG_ERROR(HW_GPU, "Invalid GS swizzle pattern offset {}", offset);
        } else {
            gs_setup.swizzle_data[offset] = value;
            gs_setup.MarkSwizzleDataDirty();
            offset++;
        }
        break;
    }

    case PICA_REG_INDEX(vs.output_mask):
        if (!regs.internal.pipeline.gs_unit_exclusive_configuration &&
            regs.internal.pipeline.use_gs == PipelineRegs::UseGS::No) {
            regs.internal.gs.output_mask.Assign(value);
        }
        break;

    case PICA_REG_INDEX(vs.bool_uniforms):
        vs_setup.WriteUniformBoolReg(regs.internal.vs.bool_uniforms.Value());
        if (!regs.internal.pipeline.gs_unit_exclusive_configuration &&
            regs.internal.pipeline.use_gs == PipelineRegs::UseGS::No) {
            gs_setup.WriteUniformBoolReg(regs.internal.vs.bool_uniforms.Value());
        }
        break;

    case PICA_REG_INDEX(vs.int_uniforms[0]):
    case PICA_REG_INDEX(vs.int_uniforms[1]):
    case PICA_REG_INDEX(vs.int_uniforms[2]):
    case PICA_REG_INDEX(vs.int_uniforms[3]): {
        const u32 index = (id - PICA_REG_INDEX(vs.int_uniforms[0]));
        vs_setup.WriteUniformIntReg(index, regs.internal.vs.GetIntUniform(index));
        if (!regs.internal.pipeline.gs_unit_exclusive_configuration &&
            regs.internal.pipeline.use_gs == PipelineRegs::UseGS::No) {
            gs_setup.WriteUniformIntReg(index, regs.internal.vs.GetIntUniform(index));
        }
        break;
    }

    case PICA_REG_INDEX(vs.uniform_setup.set_value[0]):
    case PICA_REG_INDEX(vs.uniform_setup.set_value[1]):
    case PICA_REG_INDEX(vs.uniform_setup.set_value[2]):
    case PICA_REG_INDEX(vs.uniform_setup.set_value[3]):
    case PICA_REG_INDEX(vs.uniform_setup.set_value[4]):
    case PICA_REG_INDEX(vs.uniform_setup.set_value[5]):
    case PICA_REG_INDEX(vs.uniform_setup.set_value[6]):
    case PICA_REG_INDEX(vs.uniform_setup.set_value[7]): {
        const auto index = vs_setup.WriteUniformFloatReg(regs.internal.vs, value);
        if (!regs.internal.pipeline.gs_unit_exclusive_configuration &&
            regs.internal.pipeline.use_gs == PipelineRegs::UseGS::No && index) {
            gs_setup.uniforms.f[index.value()] = vs_setup.uniforms.f[index.value()];
        }
        break;
    }

    case PICA_REG_INDEX(vs.program.set_word[0]):
    case PICA_REG_INDEX(vs.program.set_word[1]):
    case PICA_REG_INDEX(vs.program.set_word[2]):
    case PICA_REG_INDEX(vs.program.set_word[3]):
    case PICA_REG_INDEX(vs.program.set_word[4]):
    case PICA_REG_INDEX(vs.program.set_word[5]):
    case PICA_REG_INDEX(vs.program.set_word[6]):
    case PICA_REG_INDEX(vs.program.set_word[7]): {
        u32& offset = regs.internal.vs.program.offset;
        if (offset >= 512) {
            LOG_ERROR(HW_GPU, "Invalid VS program offset {}", offset);
        } else {
            vs_setup.program_code[offset] = value;
            vs_setup.MarkProgramCodeDirty();
            if (!regs.internal.pipeline.gs_unit_exclusive_configuration) {
                gs_setup.program_code[offset] = value;
                gs_setup.MarkProgramCodeDirty();
            }
            offset++;
        }
        break;
    }

    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[0]):
    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[1]):
    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[2]):
    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[3]):
    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[4]):
    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[5]):
    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[6]):
    case PICA_REG_INDEX(vs.swizzle_patterns.set_word[7]): {
        u32& offset = regs.internal.vs.swizzle_patterns.offset;
        if (offset >= vs_setup.swizzle_data.size()) {
            LOG_ERROR(HW_GPU, "Invalid VS swizzle pattern offset {}", offset);
        } else {
            vs_setup.swizzle_data[offset] = value;
            vs_setup.MarkSwizzleDataDirty();
            if (!regs.internal.pipeline.gs_unit_exclusive_configuration) {
                gs_setup.swizzle_data[offset] = value;
                gs_setup.MarkSwizzleDataDirty();
            }
            offset++;
        }
        break;
    }

    case PICA_REG_INDEX(lighting.lut_data[0]):
    case PICA_REG_INDEX(lighting.lut_data[1]):
    case PICA_REG_INDEX(lighting.lut_data[2]):
    case PICA_REG_INDEX(lighting.lut_data[3]):
    case PICA_REG_INDEX(lighting.lut_data[4]):
    case PICA_REG_INDEX(lighting.lut_data[5]):
    case PICA_REG_INDEX(lighting.lut_data[6]):
    case PICA_REG_INDEX(lighting.lut_data[7]): {
        auto& lut_config = regs.internal.lighting.lut_config;
        ASSERT_MSG(lut_config.index < 256, "lut_config.index exceeded maximum value of 255!");

        lighting.luts[lut_config.type][lut_config.index].raw = value;
        lut_config.index.Assign(lut_config.index + 1);
        break;
    }

    case PICA_REG_INDEX(texturing.fog_lut_data[0]):
    case PICA_REG_INDEX(texturing.fog_lut_data[1]):
    case PICA_REG_INDEX(texturing.fog_lut_data[2]):
    case PICA_REG_INDEX(texturing.fog_lut_data[3]):
    case PICA_REG_INDEX(texturing.fog_lut_data[4]):
    case PICA_REG_INDEX(texturing.fog_lut_data[5]):
    case PICA_REG_INDEX(texturing.fog_lut_data[6]):
    case PICA_REG_INDEX(texturing.fog_lut_data[7]): {
        fog.lut[regs.internal.texturing.fog_lut_offset % 128].raw = value;
        regs.internal.texturing.fog_lut_offset.Assign(regs.internal.texturing.fog_lut_offset + 1);
        break;
    }

    case PICA_REG_INDEX(texturing.proctex_lut_data[0]):
    case PICA_REG_INDEX(texturing.proctex_lut_data[1]):
    case PICA_REG_INDEX(texturing.proctex_lut_data[2]):
    case PICA_REG_INDEX(texturing.proctex_lut_data[3]):
    case PICA_REG_INDEX(texturing.proctex_lut_data[4]):
    case PICA_REG_INDEX(texturing.proctex_lut_data[5]):
    case PICA_REG_INDEX(texturing.proctex_lut_data[6]):
    case PICA_REG_INDEX(texturing.proctex_lut_data[7]): {
        auto& index = regs.internal.texturing.proctex_lut_config.index;

        switch (regs.internal.texturing.proctex_lut_config.ref_table.Value()) {
        case TexturingRegs::ProcTexLutTable::Noise:
            proctex.noise_table[index % proctex.noise_table.size()].raw = value;
            break;
        case TexturingRegs::ProcTexLutTable::ColorMap:
            proctex.color_map_table[index % proctex.color_map_table.size()].raw = value;
            break;
        case TexturingRegs::ProcTexLutTable::AlphaMap:
            proctex.alpha_map_table[index % proctex.alpha_map_table.size()].raw = value;
            break;
        case TexturingRegs::ProcTexLutTable::Color:
            proctex.color_table[index % proctex.color_table.size()].raw = value;
            break;
        case TexturingRegs::ProcTexLutTable::ColorDiff:
            proctex.color_diff_table[index % proctex.color_diff_table.size()].raw = value;
            break;
        }
        index.Assign(index + 1);
        break;
    }
    default:
        break;
    }

    // Notify the rasterizer an internal register was updated.
    rasterizer->NotifyPicaRegisterChanged(id);
}

void PicaCore::SubmitImmediate(u32 value) {
    // Push to word to the queue. This returns true when a full attribute is formed.
    if (!immediate.queue.Push(value)) {
        return;
    }

    constexpr std::size_t IMMEDIATE_MODE_INDEX = 0xF;

    auto& setup = regs.internal.pipeline.vs_default_attributes_setup;
    if (setup.index > IMMEDIATE_MODE_INDEX) {
        LOG_ERROR(HW_GPU, "Invalid VS default attribute index {}", setup.index);
        return;
    }

    // Retrieve the attribute and place it in the default attribute buffer.
    const auto attribute = immediate.queue.Get();
    if (setup.index < IMMEDIATE_MODE_INDEX) {
        input_default_attributes[setup.index] = attribute;
        setup.index++;
        return;
    }

    // When index is 0xF the attribute is used for immediate mode drawing.
    immediate.input_vertex[immediate.current_attribute] = attribute;
    if (immediate.current_attribute < regs.internal.pipeline.max_input_attrib_index) {
        immediate.current_attribute++;
        return;
    }

    // We formed a vertex, flush.
    DrawImmediate();
}

void PicaCore::DrawImmediate() {
    // Compile the vertex shader.
    shader_engine->SetupBatch(vs_setup, regs.internal.vs.main_offset);

    // Track vertex in the debug recorder.
    if (debug_context) {
        debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
                               std::addressof(immediate.input_vertex));
        SCOPE_EXIT(
            { debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr); });
    }

    ShaderUnit shader_unit;
    AttributeBuffer output{};

    // Invoke the vertex shader for the vertex.
    shader_unit.LoadInput(regs.internal.vs, immediate.input_vertex);
    shader_engine->Run(vs_setup, shader_unit);
    shader_unit.WriteOutput(regs.internal.vs, output);

    // Reconfigure geometry pipeline if needed.
    if (immediate.reset_geometry_pipeline) {
        geometry_pipeline.Reconfigure();
        immediate.reset_geometry_pipeline = false;
    }

    // Send to geometry pipeline.
    ASSERT(!geometry_pipeline.NeedIndexInput());
    geometry_pipeline.Setup(shader_engine.get());
    geometry_pipeline.SubmitVertex(output);

    // Flush the immediate triangle.
    rasterizer->DrawTriangles();
    immediate.current_attribute = 0;
}

void PicaCore::DrawArrays(bool is_indexed) {
    MICROPROFILE_SCOPE(GPU_Drawing);

    // Track vertex in the debug recorder.
    if (debug_context) {
        debug_context->OnEvent(DebugContext::Event::IncomingPrimitiveBatch, nullptr);
        SCOPE_EXIT(
            { debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr); });
    }

    const bool accelerate_draw = [this] {
        // Geometry shaders cannot be accelerated due to register preservation.
        if (regs.internal.pipeline.use_gs == PipelineRegs::UseGS::Yes) {
            return false;
        }

        // TODO (wwylele): for Strip/Fan topology, if the primitive assember is not restarted
        // after this draw call, the buffered vertex from this draw should "leak" to the next
        // draw, in which case we should buffer the vertex into the software primitive assember,
        // or disable accelerate draw completely. However, there is not game found yet that does
        // this, so this is left unimplemented for now. Revisit this when an issue is found in
        // games.

        bool accelerate_draw = Settings::values.use_hw_shader && primitive_assembler.IsEmpty();
        const auto topology = primitive_assembler.GetTopology();
        if (topology == PipelineRegs::TriangleTopology::Shader ||
            topology == PipelineRegs::TriangleTopology::List) {
            accelerate_draw = accelerate_draw && (regs.internal.pipeline.num_vertices % 3) == 0;
        }
        return accelerate_draw;
    }();

    // Attempt to use hardware vertex shaders if possible.
    if (accelerate_draw && rasterizer->AccelerateDrawBatch(is_indexed)) {
        return;
    }

    // We cannot accelerate the draw, so load and execute the vertex shader for each vertex.
    LoadVertices(is_indexed);

    // Draw emitted triangles.
    rasterizer->DrawTriangles();
}

void PicaCore::LoadVertices(bool is_indexed) {
    // Read and validate vertex information from the loaders
    const auto& pipeline = regs.internal.pipeline;
    const PAddr base_address = pipeline.vertex_attributes.GetPhysicalBaseAddress();
    const auto loader = VertexLoader(memory, pipeline);
    regs.internal.rasterizer.ValidateSemantics();

    // Locate index buffer.
    const auto& index_info = pipeline.index_array;
    const u8* index_address_8 = memory.GetPhysicalPointer(base_address + index_info.offset);
    const u16* index_address_16 = reinterpret_cast<const u16*>(index_address_8);
    const bool index_u16 = index_info.format != 0;

    // Simple circular-replacement vertex cache
    const std::size_t VERTEX_CACHE_SIZE = 64;
    std::array<bool, VERTEX_CACHE_SIZE> vertex_cache_valid{};
    std::array<u16, VERTEX_CACHE_SIZE> vertex_cache_ids;
    std::array<AttributeBuffer, VERTEX_CACHE_SIZE> vertex_cache;
    u32 vertex_cache_pos = 0;

    // Compile the vertex shader for this batch.
    ShaderUnit shader_unit;
    AttributeBuffer vs_output;
    shader_engine->SetupBatch(vs_setup, regs.internal.vs.main_offset);

    // Setup geometry pipeline in case we are using a geometry shader.
    geometry_pipeline.Reconfigure();
    geometry_pipeline.Setup(shader_engine.get());
    ASSERT(!geometry_pipeline.NeedIndexInput() || is_indexed);

    for (u32 index = 0; index < pipeline.num_vertices; ++index) {
        // Indexed rendering doesn't use the start offset
        const u32 vertex = is_indexed
                               ? (index_u16 ? index_address_16[index] : index_address_8[index])
                               : (index + pipeline.vertex_offset);

        bool vertex_cache_hit = false;
        if (is_indexed) {
            if (geometry_pipeline.NeedIndexInput()) {
                geometry_pipeline.SubmitIndex(vertex);
                continue;
            }

            for (u32 i = 0; i < VERTEX_CACHE_SIZE; ++i) {
                if (vertex_cache_valid[i] && vertex == vertex_cache_ids[i]) {
                    vs_output = vertex_cache[i];
                    vertex_cache_hit = true;
                    break;
                }
            }
        }

        if (!vertex_cache_hit) {
            // Initialize data for the current vertex
            AttributeBuffer input;
            loader.LoadVertex(base_address, index, vertex, input, input_default_attributes);

            // Record vertex processing to the debugger.
            if (debug_context) {
                debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
                                       std::addressof(input));
            }

            // Invoke the vertex shader for this vertex.
            shader_unit.LoadInput(regs.internal.vs, input);
            shader_engine->Run(vs_setup, shader_unit);
            shader_unit.WriteOutput(regs.internal.vs, vs_output);

            // Cache the vertex when doing indexed rendering.
            if (is_indexed) {
                vertex_cache[vertex_cache_pos] = vs_output;
                vertex_cache_valid[vertex_cache_pos] = true;
                vertex_cache_ids[vertex_cache_pos] = vertex;
                vertex_cache_pos = (vertex_cache_pos + 1) % VERTEX_CACHE_SIZE;
            }
        }

        // Send to geometry pipeline
        geometry_pipeline.SubmitVertex(vs_output);
    }
}

template <class Archive>
void PicaCore::CommandList::serialize(Archive& ar, const u32 file_version) {
    ar& addr;
    ar& length;
    ar& current_index;
    if (Archive::is_loading::value) {
        const u8* ptr = Core::System::GetInstance().Memory().GetPhysicalPointer(addr);
        head = reinterpret_cast<const u32*>(ptr);
    }
}

SERIALIZE_IMPL(PicaCore::CommandList)

} // namespace Pica